As explained on the previous page, collecting the most points in the regular season is the only way to guarantee a chance to play for the Stanley Cup Playoffs, so each team strives to maximize their regular season points each year. I set out to see if there is a formula to determine point totals given team performance. The basic R statistics package was used to create this model. 

For the gameplay component of the model, I began by testing correlation in R by observing the coefficient of determination for each component and regular season point totals. The coefficient of determination, denoted R-squared, is known as the variance in regular season points explained by each game component. For example, roughly 49% of variation in point totals can be explained by the percentages of shots saved by the teams goalie. That is a high R-squared value, as hockey is clearly a complicated game, so nearly 50% of success being determined by one player's performance is astounding. Note that a team's success cannot be explained by a single metric, because no one player can carry a team to the Stanley Cup Playoffs. A team effort is necessary.

Using vast data regarding the 2016-2017 NHL season from NHL.com and stats.HockeyAnalysis.com along with my experience watching/tracking hockey, I used the 5 team metrics that appear objectively independent and have the highest R-squared values. In other words, I selected metrics that seem to be best at predicting regular season success, while making sure not to include a single aspect of the sport multiple times. Unfortunately, independence could not be assumed, as determining distributions and associations of different metrics would be illegitimate given only a single season of data.

Of the nearly 40 metrics assessed for linear relationships, these 5 metrics stood above the rest: blocks, power play percentage, penalty kill percentage, PDO% in close 5v5 situations, and Fenwick For % in close 5v5 situations. These metrics will be thoroughly explained on the following page.